Hello Andrew, > Is MODULE_WEAKDEP new? Yes, and it has been merged into torvalds/linux.git from today: https://git.kernel.org/torvalds/c/f488790059fe7be6b2b059ddee10835b2500b603 Here the commit reference in torvalds/linux.git if you update your repo: https://github.com/torvalds/linux/commit/61842868de13aa7fd7391c626e889f4d6f1450bf I will include more references in case you want to get more information: kmod reference: https://github.com/kmod-project/kmod/commit/05828b4a6e9327a63ef94df544a042b5e9ce4fe7 kmod test-suite has also been completed: https://github.com/kmod-project/kmod/commit/d06712b51404061eef92cb275b8303814fca86ec dracut patch has also been approved: https://github.com/dracut-ng/dracut-ng/commit/8517a6be5e20f4a6d87e55fce35ee3e29e2a1150 > It seems like a "Wack a Mole" solution, which is not going to > scale. Does dracut not have a set of configuration files indicating > what modules should be included, using wildcards? If you want to have > NFS root, you need all the network drivers, and so you need all the > PHY drivers? The intention is to have a general solution not only related to the possible phy modules. That is, it is a solution for any module dependencies that are solved within the kernel but need to be known by user tools to build initramfs. We could use wildcards for some examples but it is not always easy to reference them. In addition, initramfs needs to be as small as possible so we should avoid wildcards and in this case, include the only possible phy modules (indeed not all phy's are compatible with a device). In this way, with the default behavior, initramfs would include only the drivers for the current machine and the only related phy modules. Thanks Best regards José Ignacio